Converging compute, storage and networking into a single tier sounds good at first but as the architecture scales and tries to extend into the cloud, problems arise. Most hyperconverged infrastructures are “over-converged.” In most cases, the lowest common denominator is the node, and it comes with compute, storage and networking. But what if all the organization needs are more storage capacity or more storage performance? In most cases, it has to buy more of everything to get just the specific resource it needs. The lack of granularity makes it hard to push hyperconvergence into the enterprise or extend it into the cloud.
Most hyperconverged infrastructures answer the request for one more resource by adding all resources at once. The problem is that no data center scales this way. In almost every case, a data center needs more of one resource; compute, storage performance, storage capacity or networking, not all of them at the same time.
In the traditional hyperconverged architecture, as the environment grows and needs more resources, it adds more nodes. Scale-out expansion may seem ideal but as the node count increases the data center becomes more complicated. More nodes means more network traffic, greater cluster management overhead and more data protection overhead. The problem is that most hyperconverged clusters scale more quickly than they should because they can’t add just the particular resource they need.
Ideally, each resource should scale independently from the other resources. But this independent scaling breaks the hyperconverged model and leads the customer back to traditional multi-tier, resource-specific architectures.
The Datrium Way
Datrium is a hyperconverged solution that solves the rigidity problem by tweaking the hyperconverged model. The compute tier still runs the hypervisor, virtual machines and applications. It also still runs the bulk of the storage software. This tier handles the storage performance resource. The key difference is storage capacity, which is handled via a shared storage device. The storage device provides durable capacity for less-active data and it scales independently of the compute tier. Datrium calls this design Open Convergence.
Essentially, the Datrium Open Converged architecture has two basic tiers. One for performance, both compute and storage, and another for capacity and long-term data retention. This slight tweak to the classic hyperconverged design enables customers to scale much more simply and efficiently, reducing node waste. It also allows customer to use their existing servers if they choose, mix workloads within a cluster, and converged backup cost-effectively within the primary system.
Enterprise Class Open Convergence
Thus far, Datrium’s data nodes have been disk-based. As it attracts more enterprise customers, the company is now bringing out an all-flash data node. The primary compute nodes leverage NVMe flash for optimal storage performance, alleviating the concern of network latency eliminating the NVMe performance gain. The new data nodes are standard SAS-based flash enclosures.
The value of this design is if the compute node doesn’t have the data locally, the response time from the data node is almost as fast as the compute node. The all-flash data node is also ideal for write-intensive environments. In the Datrium architecture, write IO goes directly to the data node. The all-flash data node has double the write bandwidth of the disk-based data node.
While the disk-based data node is the primary option for many customers, organizations with write-intensive workloads, like large, high transaction databases, multi-thousand seat virtual desktop infrastructures and large IOT deployments, are ideally suited for the new all-flash data node. A key use case will be Oracle and Oracle RAC deployments, which Datrium now has certification. Finally, there are clearly some data centers that have a flash-only data center goal. The all-flash data node fits in well with that vision, extending all-flash to the secondary storage environment cost effectively.
Cloud Native Open-Convergence
Datrium is also embracing the cloud with its Cloud DVX, which is a cloud-native instance of the solution running in a public cloud like Amazon. The initial use case is for the cloud is backup and recovery as-a-service for a primary Datrium DVX cluster. Since Cloud DVX is in the cloud, deduplication can be applied there across multiple DVX targets to reduce public cloud consumption costs. It can also leverage deduplication on restores, only restoring the changed segments of data. And with built-in encryption end-to-end, it avoids any VPN costs that otherwise might be needed.
As the solution matures, Datrium will add multi-site monitoring and DR automation, as well as single file restore. The company is also committed to full disaster recovery as a service features before the end of 2018.
Hyperconvergence has its place in smaller data centers, but concerns arise as it scales into the enterprise. IT planners either need to be “ok” with wasted resources or they need to stick with traditional three-tier architectures. Datrium provides middle ground by leveraging all the benefit of a hyperconverged solution without the wasted resources and complex scaling thanks to its separation of performance and capacity.